Lotus.AI.QueryGenerator (Lotus v1.0.0)

Generates query statements from natural language using ReqLLM.

Handles tool building, message handling, and response extraction. Any provider supported by ReqLLM can be used by passing a model string like "openai:gpt-4o" or "anthropic:claude-opus-4".

Summary

Types

statement_response()

Functions

generate_statement(model_string, opts)

Generate a query statement using the given model string and options.

validate_key(config)

Validate that a config contains a non-empty API key string.

Types

statement_response()

@type statement_response() :: %{
  content: String.t(),
  model: String.t(),
  variables: [map()],
  usage: %{
    prompt_tokens: non_neg_integer(),
    completion_tokens: non_neg_integer(),
    total_tokens: non_neg_integer()
  }
}

Functions

generate_statement(model_string, opts)

@spec generate_statement(
  String.t(),
  keyword()
) :: {:ok, statement_response()} | {:error, term()}

Generate a query statement using the given model string and options.

The model string should be in ReqLLM format, e.g. "openai:gpt-4o", "anthropic:claude-opus-4", "google:gemini-2.0-flash".

Options

:prompt (required) - Natural language query
:data_source (required) - Name of the data source
:api_key (required) - API key for the provider
:conversation - Conversation struct for multi-turn
:query_context - Additional context for the query
:read_only - Whether to restrict to read-only statements (default: true)
:temperature - LLM temperature (default: 0.1)
:context - Caller-supplied actor context, threaded into every query and introspection call the AI makes
:scope - Caller-supplied visibility scope, threaded the same way

validate_key(config)

@spec validate_key(map()) :: :ok | {:error, String.t()}

Validate that a config contains a non-empty API key string.